Generated using version 3.2 of the official AMS LATEX template

The East African Long Rains in Observations and Models1

Wenchang Yang ∗ Richard Seager and Mark A. Cane

Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York, USA

2

Bradfield Lyon

International Research Institute for Climate and Society

Lamont-Doherty Earth Observatory, Earth Institute at Columbia University, Palisades, New York, USA.

3

∗Corresponding author address: Wenchang Yang, Lamont-Doherty Earth Observatory, Columbia Univer-

sity, 61 Route 9W, Palisades, NY 10964.

E-mail: wyang@ldeo.columbia.edu

1

ABSTRACT4

Decadal variability of the East African precipitation during the season of Mar-Apr-May (long5

rains) is examined and the performance of a series of models in simulating the observed6

features is assessed. Observational results show that the drying trend of the long rains is7

associated with decadal natural variability associated with sea surface temperature (SST)8

variations over the Pacific Ocean. Empirical Orthogonal Function, linear regression and9

composite analyses all show the spatial pattern of the associated SST field to be La Nina-10

like. The SST-forced International Research Institute for Climate and Society (IRI) forecast11

models are able to capture the East African precipitation climatology, the decadal variability12

of the long rains and the associated SST anomaly pattern but are not consistent with the13

observation in the 1970s period. The multimodel mean of the SST-forced Coupled Model14

Intercomparison Project Phase 5 (CMIP5) AMIP experiment models capture the climatology15

and the drying trend in recent decades. The fully coupled models from the CMIP5 historical16

experiment, however, have systematic errors in simulating the East African precipitation17

climatology by underestimating the long rains while overestimating the short rains. The18

multimodel mean of the historical simulations of the long rains anomalies, which is the19

best estimate of the radiatively-forced change shows a weak wetting trend associated with20

anthropogenic forcing. The SST anomaly pattern associated with the long rains has large21

discrepancies with the observations. Our results suggest caution in projections of East22

African precipitation from CMIP5 or the relationship between the East African precipitation23

and the SST spatial pattern found in paleoclimate studies with coupled climate models.24

1

1. Introduction25

East Africa was struck by a severe drought in 2010-2011, the worst drought to occur in26

the region in the last 60 years, placing millions of people into a humanitarian crisis in this27

politically and socioeconomically vulnerable region (FEWS NET 2011; Lyon and DeWitt28

2012). The drought is characterized by failure of two consecutive rainy seasons: the short29

rains, which is the October-November-December (OND) rainy season; and the long rains,30

which is the March-April-May (MAM) rainy season (Camberlin and Philippon 2002). While31

the short rains failure was expected in the La Nina year of 2010 given the robust relationship32

with El Nino-Southern Oscillation (ENSO) (Ogallo 1988; Mason and Goddard 2001), there33

was difficulty in explaining the long rains failure since no robust forcing of the precipitation34

in this season, either oceanic or atmospheric, had been identified in past studies (Camberlin35

and Philippon 2002).36

Lyon and DeWitt (2012) considered the below-normal long rains in 2011 against a longer37

time scale background. In contrast to Williams and Funk (2011) who have suggested the38

long rains have been in a multidecadal decline, Lyon and DeWitt (2012) proposed a different39

explanation. Williams and Funk (2011) associated the drying trend with an anthropogenic-40

forced relatively rapid warming of Indian Ocean sea surface temperatures (SSTs), which they41

contend extends the warm pool and Walker circulation westward, resulting in a subsidence42

anomaly and drying over East Africa. On the contrary, Lyon and DeWitt (2012) linked the43

decline in the East African long rains with a shift to warmer SSTs over the western tropical44

Pacific and cooler SSTs over the central and eastern tropical Pacific. Most recently, Lyon45

et al. (2013) have shown the shift is part of natural multidecadal variability in the Pacific.46

Paleoclimate studies have shown that East Africa has undergone dramatic hydroclimate47

variability on multidecadal or longer time scales (Verschuren et al. 2000; Russell and John-48

son 2007) although the temporal resolution in paleoclimate records is not able to distin-49

guish long rains from short rains. By using a Monte Carlo empirical orthogonal function50

(MCEOF) approach (Anchukaitis and Tierney 2012) to synthesize different hydroclimatic51

2

proxy records, Tierney et al. (2013) not only revealed many features of East African hydro-52

climate found in previous paleoclimate studies but also identified the spatial patterns of the53

features and associated uncertainties. In addition, Tierney et al. (2013) concluded on the54

base of millennium-long control simulations from different atmosphere-ocean general circula-55

tion models (AOGCMs) and proxy records of ocean variability that hydroclimate variability56

of East Africa on multidecadal timescales is controlled by the Indian Ocean. However, this57

conclusion depends in part on the capability of these models to simulate aspects of the58

current climate of East Africa (e.g. climatology and relationship with SSTs), which needs59

further validation.60

A highly relevant issue is how the East African long rains will respond to the warming61

climate forced by anthropogenic emissions of greenhouse gases (GHGs). Climate model62

simulations assessed in the Intergovernmental Panel on Climate Change (IPCC) Fourth63

Assessment Report (AR4) or the more recent Coupled Model Intercomparison Project Phase64

5 (CMIP5, Taylor et al. 2012) suggest an expansion of the Hadley circulation (Lu et al. 2007),65

enhancement of the precipitation-minus-evaparation (P − E) pattern (i.e. wet area gets66

wetter and dry area gets drier) (Held and Soden 2006; Seager et al. 2010), weakening of the67

Walker circulation (Vecchi and Soden 2007), and regional drying trends (Seager et al. 2007;68

Hoerling et al. 2011; Kelley et al. 2012). Of these changes, the P −E pattern enhancement69

implies that the East African long rains, which are part of the Intertropical Convergence70

Zone (ITCZ), will increase. In addition, weakening of the Walker circulation also suggests a71

wetting trend over East Africa, if the relationship between the East African long rains and the72

SSTs in Lyon and DeWitt (2012) and Lyon et al. (2013) are correct. However, observational73

studies have shown that the Walker circulation seemed to be strengthening over the twentieth74

century (Cane et al. 1997; Chen et al. 2002; Compo and Sardeshmukh 2010; L’Heureux et al.75

2013). Further, the multimodel mean of the CMIP5 medium mitigation scenario (RCP4.5)76

projections shows a wetting trend over East Africa (Figure 1a) during the long rains season77

over the first half of the twenty-first century while the observations show a clear drying78

3

trend over the past three decades (Figure 1b and Figure 2a). These differences could be79

because of model error representing the forced change or because the observed record is80

dominated by natural variability. Determining whether the future in East Africa will be81

drier or wetter requirers a careful examination of the ability of models to simulate realistic82

natural variability and forced changes.83

In this paper, we will investigate the decadal variability of the East African long rains84

and its relationship with SSTs in observations and compare them with a series of model85

simulations, including the International Research Institute for Climate and Society (IRI)86

forecast models, SST-forced CMIP5 experiments (so called Atmosphere Model Intercom-87

parison Project, AMIP) and the CMIP5 historical simulations (coupled models). The key88

questions include: what is the character of decadal variability of the East African long rains89

in the observations over the past century? What is its relationship with the observed SSTs?90

Do the simulations from the SST-forced models and the fully coupled models capture the91

these features? What does that imply for the model projections of the East African long92

rains precipitation in the near future? The paper is organized as follows: data and methods93

are described in Section 2; Section 3 presents the observational results; IRI forecast model94

simulations and comparison with the observations are given in Section 4; results from CMIP595

AMIP and historical experiments are presented in Section 5 and 6, respectively; conclusions96

and a discussion are presented in Section 7.97

2. Data and methods98

For precipitation, we use the following datasets: version 5 of Global Precipitation Clima-99

tology Centre (GPCC) monthly precipitation (Rudolf et al. 2010), which is a gauge-based,100

gridded global land surface dataset for the period from 1901 to the present day; version TS101

3.1 of monthly precipitation over global land areas from Climatic Research Unit at the Uni-102

versity of East Anglia (CRU hereafter, Mitchell and Jones 2005); version 2.2 of the Global103

4

Precipitation Climatology Project (GPCP) monthly precipitation dataset from 1979 to the104

present day (Huffman et al. 2009), which combines gauge observations and satellite data into105

2.50 × 2.50 global grids; Climate Prediction Center (CPC) Merged Analysis of Precipitation106

(CMAP) (Xie and Arkin 1997), which is also monthly satellite and gauge data and covering107

from 1979 to the present.108

The observational sea surface temperature (SST) is from the version 3b of the NOAA109

National Climate Data Center (NCDC) Extended Reconstructed Sea Surface Temperature110

(ERSST) (Smith et al. 2008), which is a globally gridded monthly dataset with a spatial111

resolution of 20 by 20 and covers the period from 1854 to the present.112

Historical simulations (from 1950 to the present) from three different International Re-113

search Institute for Climate and Society (IRI) forecast models forced only with observed SST114

(http://iridl.ldeo.columbia.edu/SOURCES/.IRI/.FD/) include: ECHAM4.5 (Roeckner115

et al. 1996), ECHAM5 (Roeckner et al. 2006) and CCM3.6 (Kiehl et al. 1996). Each model116

has twenty-four ensemble members initialized with different atmospheric conditions. Both117

the ensemble mean and its range (represented as the 5th and 95th percentiles) are estimated118

for the climatology and historical time series but only the ensemble mean is used to examine119

the relationship between East African precipitation and SSTs.120

The SST-forced models used in this paper also include twelve models from the CMIP5121

AMIP experiment (Table 1). Each model has at least one ensemble member and some have122

as many as ten (e.g. CSIRO-Mk3-6-0). For each model, we only analyze the period between123

1979 to 2008 which is the longest period shared by the twelve models. We also use the fully124

coupled models from the CMIP5 historical experiment (Table 2), of which there are forty125

three in total and the period of 1850-2005 common for almost all of the models is used in126

the multimodel mean estimation. Note that the analysis of observations and the SST-forced127

models uses observed SSTs while the coupled models generate their own natural variability128

of SSTs and climate which is not expected to be synchronous with natural variability in129

nature.130

5

In this paper, the region of East Africa is represented by the land area within the box131

between 300E and 520E and 100S and 120N (red box in Figure 1). In order to reveal the SST132

anomaly pattern associated with the decadal variability the East African long rains, three133

types of analysis have been applied: Empirical Orthogonal Function (EOF) analysis (Wilks134

2011), linear regression analysis (regression of SST at each grid point on the time series of135

the East African long rains anomalies) and composite analysis (composite of SST anomalies136

over years when the East African long rains anomalies are below one standard deviation).137

In all the analyses, a nine-year running average is applied to both the precipitation and SST138

datasets before performing the analysis.139

3. Observational analysis of long rains variability and140

relation to SST variability141

Figure 2a shows the nine-year-running-averaged anomalies (relative to the 1979-2000142

base) of the East African long rains in four observational datasets: GPCC, GPCP, CMAP143

and CRU. The four datasets are consistent in showing drying over the most recent decades144

when all have data. The GPCC anomalies have a slightly smaller magnitude than the145

two satellite-gauge datasets GPCP and CMAP and the magnitude of the CRU drying146

is only about half that of GPCC. The estimated linear trends from 1983 on are −0.19147

mm/day/decade in GPCC, −0.33 mm/day/decade in GPCP, −0.35 mm/day/decade in148

CMAP and −0.08 mm/day/decade in CRU. The longer GPCC and CRU datasets suggest149

that the drying trend in the most recent decades is part of the decadal variability of climate150

over East Africa. There are at least four wet-dry cycles in the GPCC precipitation data for151

East Africa. The first one started around the beginning of the data period and ended around152

1930. The second one extended from about 1930 to the beginning of 1960s. The third one153

is relatively short, from the end of the second cycle to the middle of 1970s. The last one154

started from the end of the third cycle to the present day and probably continues. The CRU155

6

data is generally consistent with the GPCC over the whole period and shows similar decadal156

variability as in GPCC.157

Besides the decadal variability described above, there is also a small centennial-scale158

drying trend of −0.019 mm/day/decade (or −0.19 mm/day/century) in GPCC, similar to159

the time series over the same period in Figure 1b from Tierney et al. (2013). It is unclear160

whether this trend is part of natural variability on longer time scales as in Tierney et al.161

(2013) or an indication of a global warming signal as in Williams and Funk (2011) or even162

associated with data quality issues since the trend is much weaker in the CRU dataset (only163

−0.041 mm/day/century). However, if we assume the trend is due to global warming, as rep-164

resented by either global mean SST or CO2 concentration, and there is a linear relationship165

between the warming and the corresponding response in East Africa MAM precipitation,166

we can estimate the relative importance of natural and forced variability in East Africa167

MAM precipitation on decadal to centennial time scales. Figure 3a shows the nine-year-168

running-averaged East Africa MAM precipitation in GPCC before and after removing the169

postulated global warming component, as represented by global mean SST anomalies from170

ERSST. Removing the presumed global warming component decreases the magnitudes of171

wet anomalies at the beginning of the period and the dry anomalies over the last decade but172

does not change much the precipitation time series. The whole time series and the drying173

trend in recent decades are still dominated by natural variabilities. Using CO2 instead of174

global mean SST to represent the global warming forcing shows similar results (Figure 3c).175

Since there is little drying trend on the centennial scale in the CRU dataset, the global176

warming component revealed by the linear regression method is much weaker than that of177

GPCC and the dominance of natural variability is even stronger (Figure 3b and 3d).178

Lyon et al. (2013) associated the recent East African long rains drought with maps of179

global MAM precipitation and SST anomalies and attributed the recent increase in drought180

frequency to multidecadal variability of SSTs in the tropical Pacific, with cooling in the181

east and warming in the west. Can similar SST anomaly patterns be found in historical182

7

intervals when East African long rains are also in dry phases? Figure 4 shows the maps of183

MAM precipitation anomalies over land and SST anomalies in the four periods when East184

African long rains are dry. Even though each dry period has its own features, it seems the185

SST anomaly patterns over the Pacific Ocean are generally consistent over the four periods.186

Warm SST anomalies appear in the northern, northwestern and southern Pacific while cool187

SST anomalies appear over the the central to eastern Pacific. A small inconsistency is found188

over the central tropical Pacific during the period of 1945-1963 (Figure 4b) where warm SST189

anomalies appear. However, the SST anomalies over the eastern Pacific are still negative.190

Over other oceans, however, SST anomaly patterns are in general inconsistent. For example,191

the tropical Indian Ocean has warm SST anomalies during the periods of 1914-1925 (Figure192

4a) and 1994-2005 (Figure 4d) but cool anomalies during the periods of 1945-1963 (Figure193

4b) and 1968-1976 (Figure 4c). Over the Atlantic Ocean, the periods of 1914-1925 (Figure194

4a) and 1968-1976 (Figure 4c) have a pattern of cooling in the north but warming in the195

south while opposite patterns are found during the other two periods. Therefore, decadal196

variations in the East African long rains appear to be controlled by SST over the Pacific,197

consistent with the numerical experiment results in Lyon and DeWitt (2012) and Lyon198

et al. (2013), even though some studies suggest that Pacific, Indian and Atlantic Oceans all199

contribute to the decadal variability of the East African long rains (Omondi et al. 2013). The200

SST patterns over the Pacific are similar to those found in the studies of Pacific interdecadal201

climate variability such as Deser et al. (2004), although they focused on the boreal winter202

season. Along with the SST anomaly patterns over the Pacific, there are also consistent203

precipitation anomaly patterns around the global land areas (as noted by Herweijer and204

Seager 2008), including wet anomalies in the southern Africa, dry anomalies in the western205

North America and the wet-dry-wet pattern from the north to the south in South America.206

These consistent precipitation anomalies indicate that they might also be attributed to the207

same SST anomaly pattern in the Pacific as in the East African long rains case.208

To obtain a more general relationship between the East African long rains and the global209

8

SST pattern, Figure 2b shows the map of regression of SST anomaly on the negative of210

the East African precipitation anomaly during the long rains. The global average has been211

removed from the SST by regression and a nine-year running average applied to both the SST212

and the precipitation data. The regression coefficients show a La Nina-like spatial pattern,213

with negative values over the central and eastern tropical Pacific and positive values over214

the western tropical Pacific. Positive values are also found in the northern Pacific (around215

400N) and southern Pacific (around 300S). These features are consistent with the composite216

results from different dry periods in Figure 4. Choosing the years when the East African217

long rains anomalies are below one standard deviation during the whole period and making a218

composite map of the SST anomalies for these years again shows a similar pattern, confirming219

the robustness of the SST pattern (Figure 2c). The SST anomaly composite in Figure 2c220

also shows large values over the northern Atlantic but it does not mean it is associated with221

the East African long rains, as discussed in the previous paragraph.222

Can we find a mode directly from the SST data that has a similar spatial pattern to that223

shown above and at the same time correlates with the East African long rains? Figure 5224

shows the first EOF of the MAM SST after removing the global-averaged value so as to focus225

on natural variability. A nine-year running average has also been applied to emphasize the226

decadal variability. The first EOF explains 22% of the total variance and shows a La Nina-like227

spatial pattern, with negative values over the central and eastern tropical Pacific and positive228

values over the western tropical Pacific. Positive values also appear over the northern Pacific229

(centered around 400N) and the southern Pacific (centered around 300S). This pattern is very230

similar to that derived by regressing or compositing based on the precipitation data. The231

first principal component (PC1, time series from the EOF analysis) shows decadal variability232

with positive and negative anomalies lasting from years to decades. PC1 is closely correlated233

with the East African long rains anomaly (correlation coefficient of −0.62). During the recent234

drying phase and that in the 1970s, PC1 had positive values. In contrast, during the wet235

phases in the 1980s, 1930s and 1900s, PC1 had negative values. Therefore, the decadal236

9

variability of the East African long rains has a clear association with decadal SST variation237

over the Pacific, which is consistent with Lyon et al. (2013) but based on an analysis going238

further back in time.239

4. IRI forecast models240

We began by showing the East African precipitation climatology for the SST-forced241

ECHAM4.5 simulations and the GPCC observations in Figure 6. Unlike typical monsoon242

regions such as the Asian monsoon region where there is only one rainy season within a243

year and the rainy season appears in the summer months (Yang et al. 2013), there are244

primarily two rainy seasons across much of East Africa: the long rains (MAM) and the short245

rains (OND). In some areas there is a relative maximum in July and August since our East246

Africa region covers some area to the northwest where typical monsoon features emerge.247

The model simulations in general overestimate the precipitation climatology in every month248

but the overestimation is least in our months of interest, i.e. March, April and May (the249

long rains). The belt specified by the 5th and 95th percentiles of the 24 ensemble members250

is generally narrow, indicating that the simulated climatology does not differ much across251

the ensemble members. This gives us some confidence with regard to the capability of the252

model to simulate East African precipitation.253

Figure 7a then shows the East African long rains anomaly in the ECHAM4.5 model254

simulations. Comparing with the GPCC data, the ensemble mean time series captures255

the drying trend since 1990. The ensemble mean also captures the wetting trend from256

middle 1950s to the late 1960s, but does not simulate the dry anomalies in the 1950s. An257

even larger discrepancy appears in the 1970s and 1980s when there seems little relation258

between modeled and observed precipitation. The width of the belt between the 5th and259

95th percentiles exceeds the standard deviation of the ensemble mean time series, indicating260

that the modeled East African long rains are strongly influenced by internal variability in261

10

addition to the SST forced component.262

To determine the SST pattern associated with East African long rains variability, Figure263

7b shows the regression of ERSST on the ECHAM4.5 ensemble mean East African long rains264

precipitation. Comparing with the observed pattern (Figure 2b), we see that the modeled265

regression pattern captures major large scale observed features over the Pacific, although266

the modeled regression values over the northern Atlantic are much larger than the observed.267

For example, positive SST values appear in the western tropical Pacific while negative values268

lie in the central and eastern tropical Pacific. Similarly, there are also positive bands in the269

southern Pacific centered around 300S and in the northern Pacific centered around 400N.270

The composite SST anomaly map over the East African dry long rains years (Figure 7c)271

also resembles the ECHAM4.5 model captures the main features of the East African long272

rains-SST relationship.273

We also did this analysis for the two other IRI forecast models, ECHAM5 and CCM3.6274

(not shown). ECHAM5 underestimates the long rains precipitation climatology but resem-275

bles ECHAM4.5 in many of the features described above. CCM3.6 has the worst performance276

among the three models as it not only overestimates the long rains precipitation climatology277

but also has a wetting trend after 1996, which is opposite to the observed trend.278

5. CMIP5 AMIP experiment279

Figure 8 shows the East African precipitation climatology in the SST-forced CMIP5280

AMIP experiments along with the observed values from GPCC. The multimoldel ensemble281

mean agrees quite well with the GPCC climatology. The maximum model precipitation282

correctly appears in the long rain season, i.e. March, April and May. Compared with283

Figure 6, the major difference is the broad spread among ensemble members. The width284

between the 5th and 95th percentiles in the CMIP5 AMIP ensemble is much larger than for285

the ECHAM4.5 ensemble, indicating greater internal variability introduced by differences286

11

among models.287

As most models in the CMIP5 AMIP experiment only cover the period from 1979 to the288

present day, it is difficult to examine the decadal variability in these simulations. However,289

we can assess whether the simulations produce the behavior since 1979. Figure 9a shows the290

time history of the East African long rains anomaly in the CMIP5 AMIP experiment models.291

The ensemble mean shows an overall drying trend, especially after 1990. The magnitude292

of the drying trend, however, is smaller than for GPCC and occurs amidst a large spread293

among the ensemble members.294

Both the regression of SST anomaly on the East African long rains anomaly (Figure 9b)295

and the East African dry MAM years composite of SST anomalies (Figure 9c) show La Nina-296

like patterns, similar to the results in the observations and the IRI forecast models. This297

indicates that the ensemble mean of the CMIP5 AMIP experiment simulations is capable298

of capturing the relationship between the East African long rains and the SST anomaly299

pattern, although capability is limited only to recent decades.300

6. CMIP5 historical experiment301

Figure 10 shows the East African precipitation climatology in the coupled CMIP5 histor-302

ical experiments. The multimodel mean climatology does not agree with the observations; it303

is clearly worse than the SST-forced IRI forecast models and CMIP5 AMIP simulations. It304

underestimates long rains (March, April and May) and overestimates short rains (October,305

November, December). The long rains peaks during May instead of April as in observations.306

The spread is quite large. All of these problems have also been reported in CMIP3 models307

(Anyah and Qiu 2012). The drop in realism of the precipitation simulation relative to the308

SST-forced atmosphere models obviously arises in errors introduced by the simulation of309

SSTs.310

The multimodel mean among coupled models cancels out most of the internal variabilities,311

12

whether generated by internal atmospheric or coupled atmosphere-ocean variability, and312

isolates the externally forced change. The CMIP5 historical multimodel mean does not313

capture the time series of the East African long rains anomaly (Figure 11a) showing only a314

weak wetting trend after 1950. This forced wetting trend is consistent with the 21st century315

projections of the IPCC AR4 models (Figure 7a in Held and Soden 2006) and the CMIP5316

models (Figure 1a). However, as the forced response is still weak relative to the natural317

variability, it is difficult to say whether East Africa will have drier or wetter long rains in318

the near future since this will strongly depend on the future evolution of natural variability.319

Figure 11b shows the multimodel composite SST anomaly pattern for dry East African320

MAM years of the CMIP5 historical simulations. Comparing with Figure 2c, there are many321

differences from the observations. However, the CMIP5 historical simulations capture the322

observed relation with the SST gradient over the tropical Pacific Ocean, although the cool323

anomalies over the central tropical Pacific and the change in gradient are much weaker than324

observed. The significance of the composite SST anomaly pattern is indicated in Figure 11c,325

where the composite is performed for each model and the number of models with positive326

composite values at each grid point is counted. This pattern is in general consistent with327

Figure 11b with warm anomalies in the western tropical Pacific being the most robust model328

feature. The large negative values appearing over the Indian Ocean in Figures 11b and329

11c suggest that the Indian Ocean also is important for the decadal variability of the East330

African long rains in the CMIP5 coupled models. Comparing to Figure 2, we see that the331

CMIP5 coupled models overestimate the Indian Ocean’s relation to the East African long332

rains climate on these timescales.333

As the multimodel mean of the CMIP5 historical experiments is poor in capturing the334

East African long rains-SST relationship and even simulates an incorrect climatology, it335

is interesting to examine the performances of the individual models. Figure 12 shows the336

scatter plot of correlation coefficients against the differences of annual mean between the337

climatologies of the East African precipitation in the CMIP5 historical simulations and338

13

that in GPCC. The mean values among the ensemble members for each individual model339

are denoted by dots and the corresponding ranges are denoted by lines. The larger the340

correlation coefficients and the smaller the absolute values of the differences, the better341

the models’ performances. The criteria that the correlation coefficients are > 0.6 and the342

absolute values of the differences are < 0.6 mm/day (denoted by the green box) leaves343

four models that are best in simulating the East African precipitation climatolgy: CSIRO-344

Mk3-6-0, HadCM3, IPSL-CM5A-MR and bcc-csm1-1-m. The four best coupled models345

indeed do a much better job in simulating the observed climatology (Figure 13) than the346

multimodel mean (Figure 10). Models CSIRO-Mk3-6-0 and bcc-csm1-1-m have almost the347

same long rains precipitation climatology as the observed although they both overestimate348

the short rains climatology, which is the same problem as in the case of the multimodel349

mean. Models HadCM3 and IPSL-CM5A-MR capture the short rains climatology very well350

but both underestimates the long rains climatology.351

Are the four models that are best at simulating the East African precipitation climatology352

able to capture the precipitation-SST relationship in the observations as shown in Figure 2c?353

Figure 14 shows the ensemble mean of the composite SST anomalies when the East African354

long rains precipitation anomalies within each ensemble member are below one standard355

deviation. Of the four coupled models, CSIRO-Mk3-6-0, IPSL-CM5A-MR and bcc-csm1-1-m356

capture the cool anomalies over the central and eastern Pacific, although the magnitudes are357

in general smaller than observed. HadCM3 has a contrast of cool/warm to the north/south358

of the equator over the tropical Pacific, which is totally different from the observed pattern.359

The observed positive bands in the southern and northern Pacific can only be seen in the360

CSIRO-Mk3-6-0 model. The CSIRO-Mk3-6-0 model seems to be the the only model that361

captures both the East African precipitation climatology and the East African long rains362

precipitation-SST relationship in the observations.363

14

7. Conclusion and discussion364

In this paper, we have examined the decadal variability of East African precipitation365

during the long rains (Mar-May) in observations and assessed the ability of both SST-forced366

and coupled models to capture the observed features. The following conclusions have been367

reached:368

• The drying trend of the East African long rains in recent decades most likely arises369

from decadal variability of natural and is not anthropogenic origin.370

• The decadal variability of the East African long rains can be explained by the decadal371

variability of SST over the Pacific ocean. The dry phases of the East African long372

rains are associated with positive SST anomalies over the western tropical Pacific and373

negative anomalies over the central and eastern tropical Pacific (La Nina-like).374

• ECHAM4.5 and 5 from the SST-forced IRI forecast models are able to capture the375

precipitation climatology and the SST anomaly pattern associated with decadal vari-376

ability. However, not all the models perform well (e.g. CCM3.6) and the simulated377

precipitation history is in general poor.378

• The SST-forced models of the CMIP5 AMIP experiments also capture the climatology379

of the East Africa precipitation, although with varying skills. The simulated East380

African long rains anomaly, which is only available over a short period, does capture381

the recent observed drying trend in the multimodel mean and the associated SST382

anomaly pattern is consistent with the observed.383

• The multimodel mean of the fully coupled models of the CMIP5 historical experiment384

underestimates the East African long rains and overestimates the short rains with385

a considerable range of performance among the individual models. The multimodel386

mean of the precipitation anomalies shows a weak wetting trend since 1950 that is387

much smaller than the internal variability. The SST anomaly pattern associated with388

15

the dry phase of the East African long rains is associated with a SST gradient over the389

tropical Pacific, although the magnitude of this is weak compared to observations.390

• CSIRO-Mk3-6-0 is the best model among the CMIP5 historical experiment coupled391

models that captures both the East African precipitation climatology and the East392

African long rains-SST relationship in the observations.393

The 2010-2011 East African severe drought raised the question of whether it is forced by394

anthropogenic emissions of greenhouse gases or just a phase of long term natural variability395

(Lott et al. 2013). Previous studies (e.g. Held and Soden 2006) and our own estimate from396

CMIP5 (Figure 1a) show that the East African long rains are projected to get wetter under397

global warming, which is opposite to the recent drying trend. By using different methods398

of analysis and focusing on decadal variability, our study reveals that the drying trend is399

most likely part of decadal variability associated with natural SST variability mainly in the400

Pacific Ocean and with a La Nina-like pattern as in Lyon and DeWitt (2012) and Lyon et al.401

(2013). SST-forced models are able to capture the recent drying trend in the East African402

long rains and much of the decadal variability since 1950s. Fully coupled models, however,403

have systematic errors in simulating the East African precipitation climatology (i.e. under-404

estimating the long rains precipitation while overestimating the short rains precipitation)405

and give somewhat different SST patterns associated with decadal variability of the East406

African long rains. This lack of coupled model skill casts doubt on projections of future407

East African precipitation and on the use of these models to understand past variations (e.g.408

Tierney et al. 2013).409

Strong internal variability of the long rains precipitation in East Africa, and the generally410

poor ability of models to reproduce this and its atmosphere-ocean causes, means that our411

ability to predict future variations is very limited. It should not be assumed that recent412

drying trends represent an anthropogenically forced precipitation change and they will con-413

tinue. Nor should it be assumed that the model projection of wetting in response to rising414

16

greenhouse gases is correct. We are distressingly far from an adequate understanding or a415

usable ability to model climate variability and change in this socially critical region.416

Acknowledgments.417

This work was supported by NOAA award NA10OAR4310137 (Global Decadal Hydro-418

climate Variability and Change). We would like to thank Jason Smerdon, Jess Tierney and419

Yochanan Kushnir for useful discussions.420

17

421

REFERENCES422

Anchukaitis, K. J. and J. E. Tierney, 2012: Identifying coherent spatiotemporal modes in423

time-uncertain proxy paleoclimate records. Clim. Dynam., 1–16.424

Anyah, R. O. and W. Qiu, 2012: Characteristic 20th and 21st century precipitation and425

temperature patterns and changes over the Greater Horn of Africa. Int. J. Climatol.,426

32 (3), 347–363.427

Camberlin, P. and N. Philippon, 2002: The East African March-May rainy season: Associ-428

ated atmospheric dynamics and predictability over the 1968-97 period. J. Climate, 15 (9),429

1002–1019.430

Cane, M. A., A. C. Clement, A. Kaplan, Y. Kushnir, D. Pozdnyakov, R. Seager, S. E.431

Zebiak, and R. Murtugudde, 1997: Twentieth-century sea surface temperature trends.432

Science, 275 (5302), 957–960.433

Chen, J., B. E. Carlson, and A. D. Del Genio, 2002: Evidence for strengthening of the434

tropical general circulation in the 1990s. Science, 295 (5556), 838–841.435

Compo, G. P. and P. D. Sardeshmukh, 2010: Removing ENSO-related variations from the436

climate record. J. Climate, 23 (8), 1957–1978.437

Deser, C., A. S. Phillips, and J. W. Hurrell, 2004: Pacific interdecadal climate variability:438

Linkages between the tropics and the North Pacific during boreal winter since 1900. J.439

Climate, 17 (16), 3109–3124.440

FEWS NET, 2011: East Africa: Past year one of the driest on record in the eastern Horn.441

Famine Early Warning System Network Report.442

18

Held, I. and B. Soden, 2006: Robust responses of the hydrological cycle to global warming.443

J. Climate, 19 (21), 5686–5699.444

Herweijer, C. and R. Seager, 2008: The global footprint of persistent extra-tropical drought445

in the instrumental era. Int. J. Climatol., 28 (13), 1761–1774.446

Hoerling, M., et al., 2011: On North American decadal climate for 2011-20. J. Climate,447

24 (16), 4519–4528.448

Huffman, G. J., R. F. Adler, D. T. Bolvin, and G. Gu, 2009: Improving the global precipi-449

tation record: GPCP version 2.1. Geophys. Res. Lett., 36 (17).450

Kelley, C., M. Ting, R. Seager, and Y. Kushnir, 2012: Mediterranean precipitation clima-451

tology, seasonal cycle, and trend as simulated by CMIP5. Geophys. Res. Lett., 39 (21).452

Kiehl, J. T., J. J. Hack, G. B. Bonan, and B. Boville, 1996: Description of the NCAR453

community climate model(CCM 3). Tech Note NCAR/TN-420+STR.454

L’Heureux, M. L., S. Lee, and B. Lyon, 2013: Recent multidecadal strengthening of the455

Walker circulation across the tropical Pacific. Nature Climate Change, 3 (6), 571–576.456

Lott, F. C., N. Christidis, and P. A. Stott, 2013: Can the 2011 East African drought be457

attributed to human-induced climate change? Geophys. Res. Lett., 1–5.458

Lu, J., G. Vecchi, and T. Reichler, 2007: Expansion of the Hadley cell under global warming.459

Geophys. Res. Lett, 34 (10.1029).460

Lyon, B., A. Barnston, and D. G. DeWitt, 2013: Tropical Pacific forcing of a 1998-99 climate461

shift: Observational analysis and climate model results for the boreal spring season. Clim.462

Dynam., (in press).463

Lyon, B. and D. G. DeWitt, 2012: A recent and abrupt decline in the East African long464

rains. Geophys. Res. Lett., 39 (2).465

19

Mason, S. J. and L. Goddard, 2001: Probabilistic precipitation anomalies associated with466

ENSO. Bull. Amer. Meteor. Soc., 82 (4), 619–638.467

Mitchell, T. D. and P. D. Jones, 2005: An improved method of constructing a database468

of monthly climate observations and associated high-resolution grids. Int. J. Climatol.,469

25 (6), 693–712.470

Ogallo, L., 1988: Relationships between seasonal rainfall in East Africa and the southern471

oscillation. J. Climatol., 8 (1), 31–43.472

Omondi, P., L. A. Ogallo, R. Anyah, J. M. Muthama, and J. Ininda, 2013: Linkages between473

global sea surface temperatures and decadal rainfall variability over Eastern Africa region.474

Int. J. Climatol., 33 (8), 2082–2104.475

Roeckner, E., et al., 1996: The atmospheric general circulation model ECHAM-4: Model476

description and simulation of present-day climate. Max-Planck-Institut furr Meteorologie.477

Roeckner, E., et al., 2006: Sensitivity of simulated climate to horizontal and vertical resolu-478

tion in the ECHAM5 atmosphere model. J. Climate, 19 (16), 3771–3791.479

Rudolf, B., A. Becker, U. Schneider, A. Meyer-Christoffer, M. Ziese, et al., 2010: GPCC480

status report December 2010. GPCC, December.481

Russell, J. and T. Johnson, 2007: Little Ice Age drought in equatorial Africa: Intertropi-482

cal Convergence Zone migrations and El Nino–Southern Oscillation variability. Geology,483

35 (1), 21–24.484

Seager, R., N. Naik, and G. A. Vecchi, 2010: Thermodynamic and dynamic mechanisms for485

large-scale changes in the hydrological cycle in response to global warming. J. Climate,486

23 (17), 4651–4668.487

Seager, R., et al., 2007: Model projections of an imminent transition to a more arid climate488

in southwestern North America. Science, 316 (5828), 1181.489

20

Smith, T. M., R. W. Reynolds, T. C. Peterson, and J. Lawrimore, 2008: Improvements490

to NOAA’s historical merged land-ocean surface temperature analysis (1880-2006). J.491

Climate, 21 (10), 2283–2296.492

Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the493

experiment design. Bull. Amer. Meteor. Soc., 93 (4), 485–498.494

Tierney, J. E., J. E. Smerdon, K. J. Anchukaitis, and R. Seager, 2013: Multidecadal vari-495

ability in East African hydroclimate controlled by the Indian Ocean. Nature, 493 (7432),496

389–392.497

Vecchi, G. A. and B. J. Soden, 2007: Global warming and the weakening of the tropical498

circulation. J. Climate, 20 (17), 4316–4340.499

Verschuren, D., K. R. Laird, and B. F. Cumming, 2000: Rainfall and drought in equatorial500

east Africa during the past 1,100 years. Nature, 403 (6768), 410–414.501

Wilks, D. S., 2011: Statistical methods in the atmospheric sciences, Vol. 100. Academic press.502

Williams, A. P. and C. Funk, 2011: A westward extension of the warm pool leads to a west-503

ward extension of the Walker circulation, drying eastern Africa. Clim. Dynam., 37 (11-504

12), 2417–2435.505

Xie, P. and P. A. Arkin, 1997: Global precipitation: A 17-year monthly analysis based on506

gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor.507

Soc., 78 (11), 2539–2558.508

Yang, W., R. Seager, and M. A. Cane, 2013: Zonal momentum balance in the tropical509

atmospheric circulation during the global monsoon mature months. J. Atmos. Sci., 70 (2),510

583–599.511

21

List of Tables512

1 Models of the CMIP5 AMIP experiment used in our analysis. 23513

2 Models of the CMIP5 historical experiment used in our analysis. 24514

22

Table 1. Models of the CMIP5 AMIP experiment used in our analysis.

Index Name Time Span Ensemble #1 CanAM4 1950-2009 42 CNRM-CM5 1979-2008 13 CSIRO-Mk3-6-0 1979-2009 104 GFDL-HIRAM-C180 1979-2008 35 GFDL-HIRAM-C360 1979-2008 26 GISS-E2-R 1880-2010 17 HadGEM2-A 1979-2008 18 inmcm4 1979-2008 19 IPSL-CM5A-LR 1979-2009 510 MPI-ESM-LR 1979-2008 311 MRI-AGCM3-2H 1979-2008 112 NorESM1-M 1979-2008 3

23

Table 2. Models of the CMIP5 historical experiment used in our analysis.

Index Name Time Span Ensemble #1 ACCESS1-0 1850-2005 12 ACCESS1-3 1850-2005 13 bcc-csm1-1 1850-2012 34 bcc-csm1-1-m 1850-2012 35 BNU-ESM 1850-2005 16 CanCM4 1961-2005 107 CanESM2 1850-2005 58 CCSM4 1850-2005 69 CESM1-BGC 1850-2005 110 CESM1-CAM5 1850-2005 311 CESM1-CAM5-1-FV2 1850-2005 412 CESM1-FASTCHEM 1850-2005 313 CESM1-WACCM 1955-2005 414 CMCC-CESM 1850-2005 115 CMCC-CM 1850-2005 116 CMCC-CMS 1850-2005 117 CNRM-CM5 1850-2005 1018 CSIRO-Mk3-6-0 1850-2005 1019 FGOALS-g2 1900-2005 520 FGOALS-s2 1850-2005 321 FIO-ESM 1850-2005 322 GFDL-CM3 1860-2005 523 GFDL-ESM2G 1861-2005 324 GFDL-ESM2M 1861-2005 125 GISS-E2-H 1850-2005 526 GISS-E2-R 1850-2005 627 HadCM3 1860-2005 1028 HadGEM2-CC 1960-2004 329 HadGEM2-ES 1860-2004 430 inmcm4 1850-2005 131 IPSL-CM5A-LR 1850-2005 532 IPSL-CM5A-MR 1850-2005 133 IPSL-CM5B-LR 1850-2005 134 MIROC-ESM 1850-2005 335 MIROC-ESM-CHEM 1850-2005 136 MIROC4h 1950-2005 337 MIROC5 1850-2005 438 MPI-ESM-LR 1850-2005 339 MPI-ESM-MR 1850-2005 340 MPI-ESM-P 1850-2005 241 MRI-CGCM3 1850-2005 342 NorESM1-M 1850-2005 343 NorESM1-ME 1850-2005 1

24

List of Figures515

1 (a): MAM precipitation rate difference between 2046-2065 and 2006-2025 from516

CMIP5 RCP4.5 projection multimodel mean. (b): MAM precipitation rate517

linear trend over the period of 1979-2008 from GPCP dataset. Red boxes in518

the two panels denote the East Africa area between 300E and 520E and 100S519

and 120N. 28520

2 (a): East Africa (30E-52E, 10S-12N) MAM precipitation rate anomaly relative521

to the 1979-2000 base in observations. Nine-year running average has been522

applied to all the time series. (b): Regression of SST anomaly on the negative523

of East Arica precipitation rate anomaly (i.e. b as in ssta = −b × pra) in524

MAM over the period of 1901-2009. (c): Composite of MAM SST anomaly525

from years when East Africa MAM precipitation anomaly is below its one526

standard deviation. In both (b) and (c), The global average SST has been527

removed from the SST data by linear regression and a nine-year running528

average has been applied to both the SST and the precipitation. The SST529

and precipitation data are from ERSST and GPCC respectively. 29530

3 East Africa MAM precipitation anomalies (gray lines) and their residuals531

(black lines) after removing the global warming signals (red lines). Dashed532

lines are the corresponding linear trends (mm/day/century). (a) and (c) use533

the GPCC dataset while (b) and (d) use the CRU precipitation dataset.534

Global mean of MAM ERSST is used as the global warming signal in (a)535

and (b) while annual mean CO2 concentration is used in (c) and (d). All time536

series have been low-pass filtered by applying a nine-year running average. 30537

25

4 Composite of MAM GPCC precipitation rate anomalies over land (brown-538

green colors) and MAM ERSST anomalies (blue-red colors). Global mean539

has been removed from the SST by linear regression and a nine-year running540

average has been applied to both the SST and precipitation datasets before541

performing the composite analysis. 31542

5 The first EOF of the ERSST in MAM after removing the global mean SST543

and performing a 9-year running average. (a): The first EOF spatial pattern.544

(b): The first principal component (time series from the EOF analysis, lighter545

gray area plot, unit K) and East Africa MAM precipitationanomaly (gray546

bars, unit mm/day). 32547

6 East Africa (30E-52E, 10S-12N) precipitation rate climatology in GPCC and548

ECHAM4.5 model simulations estimated over the period of 1979-2005. The549

gray area indicates the range between the 95th and the 5th percentiles of the550

24 ensemble members. 33551

7 Same as Figure 2 but for ECHAM4.5 instead of observations. The gray area552

in (a) indicates the range between the 95th and the 5th percentiles of the 24553

ensemble members. The ensemble mean time series in (a) is used in (b) and554

(c). 34555

8 Same as Figure 6 but for CMIP5 AMIP multimodel runs instead of ECHAM4.5. 35556

9 Same as Figure 7 but for CMIP5 AMIP multimodel runs instead of ECHAM4.5. 36557

10 Same as Figure 6 but for CMIP5 historical multimodel runs instead of ECHAM4.5. 37558

11 (a) and (b) are same as Figure 7a and 7c respectively, but for CMIP5 historical559

multimodel runs instead of ECHAM4.5. (c): Number of models with positive560

values in the composite analysis. 38561

26

12 Scatter plot of correlation coefficients against annual mean differences between562

the climatology of East Africa precipitation in CMIP5 historical experiment563

models and that in GPCC. Blue dots denote the ensemble mean values of564

each model and the blue lines represent the range of ensemble members of565

each model. Numbers to the bottom-right of the blue dots are the indices of566

the forty-three CMIP5 historical experiment models. Models within the green567

box are the four best models based on the two scores of correlation coefficients568

and the annual mean difference. 39569

13 Same as Figure 10 but for the four best models. 40570

14 Same as Figure 11b but for the four best models. 41571

27

Fig. 1. (a): MAM precipitation rate difference between 2046-2065 and 2006-2025 fromCMIP5 RCP4.5 projection multimodel mean. (b): MAM precipitation rate linear trend overthe period of 1979-2008 from GPCP dataset. Red boxes in the two panels denote the EastAfrica area between 300E and 520E and 100S and 120N.

28

Fig. 2. (a): East Africa (30E-52E, 10S-12N) MAM precipitation rate anomaly relative tothe 1979-2000 base in observations. Nine-year running average has been applied to all thetime series. (b): Regression of SST anomaly on the negative of East Arica precipitation rateanomaly (i.e. b as in ssta = −b×pra) in MAM over the period of 1901-2009. (c): Compositeof MAM SST anomaly from years when East Africa MAM precipitation anomaly is belowits one standard deviation. In both (b) and (c), The global average SST has been removedfrom the SST data by linear regression and a nine-year running average has been applied toboth the SST and the precipitation. The SST and precipitation data are from ERSST andGPCC respectively.

29

1900 1920 1940 1960 1980 2000 2020−0.4

−0.2

0

0.2

0.4

−0.19

−0.011

Pre

cip

itation (

mm

/day)

(a): GPCC and ERSST

1900 1920 1940 1960 1980 2000 2020 1900 1920 1940 1960 1980 2000 2020

−0.041 0.0021

(b): CRU and ERSST

1900 1920 1940 1960 1980 2000 2020−0.4

−0.2

0

0.2

0.4

Glo

bal m

ean S

ST

anom

aly

(K

)

1900 1920 1940 1960 1980 2000 2020−0.4

−0.2

0

0.2

0.4

−0.19

0.0045

Years

Pre

cip

itation (

mm

/day)

(c): GPCC and CO2

1900 1920 1940 1960 1980 2000 2020 1900 1920 1940 1960 1980 2000 2020

−0.041 0.027

Years1900 1920 1940 1960 1980 2000 2020

240

280

320

360

400

CO

2 (

ppm

)

(d): CRU and CO2

Fig. 3. East Africa MAM precipitation anomalies (gray lines) and their residuals (blacklines) after removing the global warming signals (red lines). Dashed lines are the corre-sponding linear trends (mm/day/century). (a) and (c) use the GPCC dataset while (b) and(d) use the CRU precipitation dataset. Global mean of MAM ERSST is used as the globalwarming signal in (a) and (b) while annual mean CO2 concentration is used in (c) and (d).All time series have been low-pass filtered by applying a nine-year running average.

30

Fig. 4. Composite of MAM GPCC precipitation rate anomalies over land (brown-greencolors) and MAM ERSST anomalies (blue-red colors). Global mean has been removed fromthe SST by linear regression and a nine-year running average has been applied to both theSST and precipitation datasets before performing the composite analysis.

31

Fig. 5. The first EOF of the ERSST in MAM after removing the global mean SST andperforming a 9-year running average. (a): The first EOF spatial pattern. (b): The firstprincipal component (time series from the EOF analysis, lighter gray area plot, unit K) andEast Africa MAM precipitationanomaly (gray bars, unit mm/day).

32

EA

pre

cip

ita

tio

n c

lima

tolo

gy(m

m/d

ay)

Month

1 2 3 4 5 6 7 8 9 10 11 120

0.5

1

1.5

2

2.5

3

3.5

4

GPCC

ECHAM4.5 ensemble mean

Fig. 6. East Africa (30E-52E, 10S-12N) precipitation rate climatology in GPCC andECHAM4.5 model simulations estimated over the period of 1979-2005. The gray area indi-cates the range between the 95th and the 5th percentiles of the 24 ensemble members.

33

Fig. 7. Same as Figure 2 but for ECHAM4.5 instead of observations. The gray area in (a)indicates the range between the 95th and the 5th percentiles of the 24 ensemble members.The ensemble mean time series in (a) is used in (b) and (c).

34

EA

MA

M p

r clim

ato

log

y(m

m/d

ay)

Month

2 4 6 8 10 120

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

GPCC

CMIP5 AMIP multiple−model mean

Fig. 8. Same as Figure 6 but for CMIP5 AMIP multimodel runs instead of ECHAM4.5.

35

Fig. 9. Same as Figure 7 but for CMIP5 AMIP multimodel runs instead of ECHAM4.5.

36

EA

pr

clim

ato

log

y(m

m/d

ay)

month

2 4 6 8 10 120

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

GPCC

CMIP5 historical multiple−model mean

Fig. 10. Same as Figure 6 but for CMIP5 historical multimodel runs instead of ECHAM4.5.

37

Fig. 11. (a) and (b) are same as Figure 7a and 7c respectively, but for CMIP5 historicalmultimodel runs instead of ECHAM4.5. (c): Number of models with positive values in thecomposite analysis.

38

−0.4 −0.2 0 0.2 0.4 0.6 0.8 1−1

−0.5

0

0.5

1

1.5

2

2.5

1

2

3

4 bcc−csm1−1−m

5

6

7

8

9 10

11

12

13

14

15

16

17

18 CSIR

O−M

k3−6−0

19

20

21 22

23 24

25

26

27 HadC

M3

28 29

30

31

32 IPSL−CM

5A−MR

33

34 35

36

37

38

39

40

41

42

43

Correlation coefficients

CM

IP5_his

torical −

GP

CC

(m

m/d

ay)

GPCC = 2.1mm/day

Fig. 12. Scatter plot of correlation coefficients against annual mean differences between theclimatology of East Africa precipitation in CMIP5 historical experiment models and that inGPCC. Blue dots denote the ensemble mean values of each model and the blue lines representthe range of ensemble members of each model. Numbers to the bottom-right of the bluedots are the indices of the forty-three CMIP5 historical experiment models. Models withinthe green box are the four best models based on the two scores of correlation coefficientsand the annual mean difference.

39

1 2 3 4 5 6 7 8 9 10 11 120

1

2

3

4

(a): CSIRO−Mk3−6−0

1 2 3 4 5 6 7 8 9 10 11 120

1

2

3

4

(b): HadCM3

1 2 3 4 5 6 7 8 9 10 11 120

1

2

3

4

EA

pre

cip

itation c

limato

logy(m

m/d

ay)

(c): IPSL−CM5A−MR

1 2 3 4 5 6 7 8 9 10 11 120

1

2

3

4

Month

(d): bcc−csm1−1−m

Fig. 13. Same as Figure 10 but for the four best models.

40

Fig. 14. Same as Figure 11b but for the four best models.

41